【摘要】 为提高相关滤波类算法的性能,提出一种基于通道选择与目标重检的跟踪算法。通过计算每个卷积通道的特征均值和方差,选取符合条件的通道训练滤波器;根据滤波响应均值找出跟踪失败的帧,扩大搜索范围后使用最优模板再次检测丢失的目标位置;通过调整搜索框的比例来训练尺度滤波器,使用平均峰值相关能量约束模板更新。实验结果表明,所提算法的准确率相较于其它算法有明显提高,有效解决了跟踪过程中的形变、遮挡和尺度变化等问题。更多还原

【Abstract】 To improve the accuracy and success rate of kernel correlation filter-based algorithms,a tracking algorithm based on channel selection and target re-detection was proposed.By calculating the feature mean and variance of each convolution channel,a channel that met the criteria to train the filter was selected.The track failed frames were found out based on the response mean value,and the optimal template was used to detect the object location again after expanding the searching range.The ratio of the searching box was adjusted to train the location filter.The template updating was restrained using mean peak value correlation energy.Experimental results show that the accuracy of the proposed algorithm is significantly improved compared to that of other algorithms,and it effectively solves the problems of deformation,occlusion,and scale changes in the tracking process.更多还原